« Bigger fish to fry | Main | The Honourable Mustache from Toronto-Danforth »

January 16, 2006


Scott Gilbreath of a new-to-me blog called Magic Statistics has "A primer on crime statistics". It's a very interesting and informative post except inasmuch as it inexplicably goes out of its way to misrepresent a few things I said in my post about David Frum's dubious American vs. Canadian crime statistics.

Feature this:

Mr Selley, I think, goes too far, however, when he rejects outright the use of victimisation surveys to measure crime:
Frum isn't actually talking about the crime rate but about the crime victimization rate, which relies on citizens self-reporting whether they were victims of crime in the preceding year. (That's just what it sounds like - the crime, if indeed there was one, needn't even have been reported to police to qualify.)

Does Mr Selley think that unreported crime is ipso facto unimportant, if not imaginary? Is he assuming that very few, if any, serious crimes go unreported? Even sexual assaults? His reference to the crime, if indeed there was one implies that he believes survey respondents systematically fabricate incidents when asked about their personal experiences with crime. At the very least, he seems to think that society need concern itself only with those crimes that are actually reported to the police.

I never really considered whether rates of reported crime victimization were important or not. As Gilbreath says, I'm sure the various agencies in question wouldn't keep the statistics if they weren't useful. I was concerned only with incorrectly calling a "crime victimization rate" a "crime rate", as Frum did, and with the extremely limited comparability of Canada's rate to the United States'.

Do I believe that "respondents systematically fabricate incidents when asked about their personal experiences"? Uh, no. Well, maybe. Look, actually, I have no idea. David Frum came out swinging about a crime rate, is all, and then it turned out that he was talking about the rate at which people report things they perceive to have been crimes. Logically, some of them weren't crimes. And as Gilbreath points out, it makes no sense that the American violent crime rate would be four times the Canadian one. So there's something wrong with those statistics beyond the fact that Frum was misrepresenting what exactly they are. My post, as I think was rather clear, was concerned with exploring the 50% figure and nothing else.

Then there's this:

John Lott's oft-cited article posted at National Review Online last August is completely erroneous… (Note, however, that, contrary to Chris Selley’s implication, this was not the source of David Frum’s statistic.)

I implied no such thing. John Lott's name doesn't even appear in my post. I linked to the article in question as an example of how prevalent the "50% greater" stat is, and meant nothing more by it than that.

(h/t Shaidle)

Posted by Chris Selley at January 16, 2006 11:23 PM

Trackback Pings

TrackBack URL for this entry:


So, basically, Canadians are more sensitive to criminality than Americans, thus more of them feel they have been the victim of crime. Except, nobody asked me, or anyone I know. The statistic is an extrapolation based on a sample. (Is the sampling consistent accross Canada and throughout the US?)

If I was murdered, or my car was stolen, that would be recorded. Thus murder rates carry the weight of objective facts, or as close as a statistic can get to one anyway.

It is like comparing an opinion poll with an election result.

Posted by: wsam at January 17, 2006 10:27 AM


To deal with your second objection first: it appeared to me that you implicitly connected John Lott's statistic with Frum's, but if you say you didn't intend that, I won't quibble. That's good enough for me. My post has been amended accordingly.

As for the other, well, it kinda looks like you might be agreeing with my interpretation regarding your view of crime that is not reported to police. In any case, the dichotomy you see between "crime rate" and "crime victimisation rate" is, in my opinion, wayward. Perhaps it wasn't clear enough, but one of the points of my post is that there is no such thing as "the crime rate". There are crime rates based on police reports and there are crime rates based on victim surveys. Both have their legitmate uses, advantages, and disadvantages. But neither one is "the" unqualified crime rate.


Posted by: StatGuy at January 17, 2006 10:35 AM

wsam posted a comment while I was writing my first. Your stolen car would be recorded in police-repoted crime statistics if you reported it to the police. As for homicide, it's a little hard to question victims about their crime experience, so homicide cannot be included in crime victimisation surveys. One has to rely on police reports for homicide rates which, as you say, we have reasonable confidence would be complete.

According to the ICVS (see page 2 of this pdf document), 91% of car thefts are reported to the police, but only 28% of sexual assaults. (Those are averages for the 17 developed countries surveyed.)

Yes, those are estimates based on sample surveys, but the ICVS administers the same questionnaires and employs the same methodology in all countries surveyed. Most of the charts in the ICVS reports show 95% confidence intervals for the estimates, so statistical significance can be assessed. See, e.g., the chart on page 3 of the linked document.

Actually, it's not like comparing an election result with an opinion poll. Police reports do not include every criminal incident because not all are reported to them. Sexual assaults, it seems to me, are very serious, but typically less than half are reported.

Posted by: StatGuy at January 17, 2006 10:54 AM

You've just explained why crime victimization surveys are bullshit.

"91% of car thefts are reported to the police, but only 28% of sexual assaults."

Murder and car theft are the only crime stats worth measuring and therefore are the only ones worth talking about. Because they are the only means by which we gain a semi-accurate picture of what is actually going on out there.

Just like comparing an opinion poll to an election result. One measures reality (the number of people who actually, in reality voted for a particular political party), the other projects out from a random sample (the number of people who said they would, sometime in the future, vote for a particular political party) to create a hypothesis about what might be happening.

Reality. Not reality.

Posted by: wsam at January 17, 2006 12:24 PM

FWIW, I hope you might want to consider what motivation there might be for people to lie to crime victimisation surveyors about their crime experience. Such surveys are confidential and random, and the surveyors are anonymous to the respondents. Those who conduct the survey are forbidden to tell anyone outside the surveying office what they are told by respondents. (I know: I've been involved in lots of them.) Canada's Statistics Act and similar provincial and territorial legislation prescribe jail time for surveyors, supervisors, and any other agency employees who violate respondent confidentiality.

Suppose a respondent tells a surveyor that she has been sexually assaulted. No one, beyond individuals directly involved in conducting and overseeing the survey, is ever going to know. If someone really wanted to capitalise on a false accusation of sexual assault, she would go to the cops or a lawyer or the press. If a respondent fabricates an incident to a crime victimisation surveyor, no information--neither the respondent's name nor any particulars of the alleged offence--will ever be passed on to police or any other legal authorities. There's absolutely nothing to gain.

Posted by: StatGuy at January 17, 2006 04:28 PM